Research On Statistical Modeling And Path Optimization Of Low-Latency And High-reliability Vehicular Networks

At present,there are some key problems in the field of unmanned driving,such as the dead angle of single vehicle sensor and the distance limit of short distance direct communication.The Cellular Vehicle to Everything(C-V2X)has become an effective way to solve these problems.By receiving real-time information from the Internet and other vehicles,connected vehicles can significantly improve the perception ability and range of the environment.However,driverless vehicles still have the problem of limited computing power,which can’t meet the business requirements of high computing.Therefore,the connected smart vehicles supporting cloud computing solves the problem of insufficient computing power of single vehicle.However,the Internet of Vehicles which only supports cloud computing has a large communication latency,which can’t guarantee the low latency and high reliability of the Internet of Vehicles business.As a supplement to cloud computing,the Internet of Vehicles system supporting fog computing can effectively reduce the network latency and significantly improve the reliability through the nearby communication and services.Existing Internet of Vehicles research work is mainly focused on analyzing the latency of queuing and workload processing,while ignoring the wireless access latency between the vehicle and the fog / cloud server,which is one of the main factors causing endto-end latency.In view of this,this paper focuses on the wireless access latency between networked vehicles and cloud data centers and fog nodes,and proposes a latency performance analysis and path selection optimization method based on fog computing.First of all,this paper conducted a four-month latency test and research on the campus of Huazhong University of Science and Technology.By installing the self-developed test software on the vehicle smart phone,we tested the wireless access latency between the school bus and the fog / cloud server by using the commercial cellular wireless communication system,and collected the Round-Trip Time(RTT)between the vehicle and the fog / cloud server and the vehicle status regularly State data.Then,an empirical spatial statistical model is established by using the latency data set to represent the spatial variation of waiting time on the main driving routes.Because latency is highly dynamic between continuous time slots and adjacent location points,our main goal is to maximize the service confidence,that is,the probability that Internet of Vehicles service can be satisfied within its tolerable latency threshold.The statistical data show that the RTT empirical probability distribution function(PDF)has a strong spatial correlation.In view of this,this paper introduces an improved method based on K-means to classify the empirical PDF of different positions in the whole research area into a limited number of probability distribution,and make the area division.Finally,considering the primary security problem in the driverless Internet of vehicles,this paper focuses on the path planning of vehicle with the goal of optimizing the average confidence of RTT and propose a path planning algorithm based on deep Q-learning network(DQN).Simulation results show that the algorithm can significantly improve the average confidence of RTT.